To quantitatively examine the efficacy of vegetation restoration in drylands globally.
Study-level viz to document patterns in exclusions primarily and the relatie frequenices, at the study level, of major categories of evidence.
#study data####
library(tidyverse)
studies <- read_csv("data/studies.csv")
studies
## # A tibble: 278 x 18
## ID title technique data region exclude rationale observations
## <dbl> <chr> <chr> <chr> <chr> <chr> <chr> <chr>
## 1 152 Shor… seeding,… expe… Africa no <NA> <NA>
## 2 180 Rest… chemical… App.… Africa no <NA> <NA>
## 3 229 Infl… soil see… fiel… Africa no <NA> <NA>
## 4 230 Acti… planting fiel… Africa no <NA> <NA>
## 5 255 The … grazing … fiel… Africa no <NA> <NA>
## 6 262 Reve… seeding,… eper… Africa no <NA> <NA>
## 7 263 The … phytogen… fiel… Africa no <NA> <NA>
## 8 264 Eval… seeding,… fiel… Africa no <NA> <NA>
## 9 271 Patc… natural … fiel… Africa no <NA> <NA>
## 10 4 Fact… natural … App.… Africa no <NA> <NA>
## # ... with 268 more rows, and 10 more variables: disturbance <chr>,
## # system <chr>, goal <chr>, intervention <chr>, paradigm <chr>,
## # grazing <chr>, hypothesis <chr>, soil <chr>, benchmark <chr>,
## # notes <chr>
#quick look at rationale needed
exclusions <- studies %>%
filter(exclude == "yes")
#quick look at studies with paradigms
evidence <- studies %>%
filter(exclude == "no")
#library(skimr)
#skim(evidence)
#study-level viz#####
#exclusions
ggplot(exclusions, aes(rationale, fill = region)) +
geom_bar() +
coord_flip() +
labs(x = "rational for exclusion", y = "frequency") +
scale_fill_brewer(palette = "Paired")
ggplot(evidence, aes(disturbance, fill = paradigm)) +
geom_bar(na.rm = TRUE) +
coord_flip() +
scale_fill_brewer(palette = "Paired") +
labs(y = "frequency")
ggplot(evidence, aes(region, fill = paradigm)) +
geom_bar(na.rm = TRUE) +
coord_flip() +
scale_fill_brewer(palette = "Paired") +
labs(y = "frequency")
ggplot(evidence, aes(data, fill = paradigm)) +
geom_bar(na.rm = TRUE) +
coord_flip() +
scale_fill_brewer(palette = "Paired") +
labs(y = "frequency")
ggplot(evidence, aes(system, fill = paradigm)) +
geom_bar(na.rm = TRUE) +
coord_flip() +
scale_fill_brewer(palette = "Paired") +
labs(y = "frequency")
ggplot(evidence, aes(goal, fill = paradigm)) +
geom_bar(na.rm = TRUE) +
coord_flip() +
scale_fill_brewer(palette = "Paired") +
labs(x = "outcome", y = "frequency")
#step 1 models####
#paradigm
derived.evidence <- evidence %>%
group_by(technique, data, region, disturbance, goal, paradigm) %>% summarise(n = n())
#active-passive split
m <- glm(n~paradigm, family = poisson, derived.evidence)
anova(m, test="Chisq")
## Analysis of Deviance Table
##
## Model: poisson, link: log
##
## Response: n
##
## Terms added sequentially (first to last)
##
##
## Df Deviance Resid. Df Resid. Dev Pr(>Chi)
## NULL 167 9.9147
## paradigm 1 0.045115 166 9.8696 0.8318
#region
m1 <- glm(n~paradigm*region, family = poisson, derived.evidence)
#m1
#summary(m1)
anova(m1, test="Chisq")
## Analysis of Deviance Table
##
## Model: poisson, link: log
##
## Response: n
##
## Terms added sequentially (first to last)
##
##
## Df Deviance Resid. Df Resid. Dev Pr(>Chi)
## NULL 167 9.9147
## paradigm 1 0.045115 166 9.8696 0.8318
## region 6 0.301367 160 9.5682 0.9995
## paradigm:region 6 0.213627 154 9.3546 0.9998
#outcome
m2 <- glm(n~paradigm*goal, family = poisson, derived.evidence)
#m1
#summary(m1)
anova(m2, test="Chisq")
## Analysis of Deviance Table
##
## Model: poisson, link: log
##
## Response: n
##
## Terms added sequentially (first to last)
##
##
## Df Deviance Resid. Df Resid. Dev Pr(>Chi)
## NULL 167 9.9147
## paradigm 1 0.045115 166 9.8696 0.8318
## goal 6 0.240941 160 9.6287 0.9997
## paradigm:goal 4 0.301480 156 9.3272 0.9897
#even split between active and passive evidence by all key categories
A summary of sort process using PRISMA.
library(PRISMAstatement)
prisma(found = 1504,
found_other = 5,
no_dupes = 1039,
screened = 1039,
screen_exclusions = 861,
full_text = 178,
full_text_exclusions = 100,
qualitative = 78,
quantitative = 78,
width = 800, height = 800)
Check data and calculate necessary measures.
data <- read_csv("data/data.csv")
data <- data %>%
mutate(lrr = log(mean.t/mean.c), rii = ((mean.t-mean.c)/(mean.t + mean.c)), var.es = ((sd.t^2/n.t*mean.t^2) + (sd.c^2/n.c*mean.c^2)))
#data
#consider adding some other effect size measures and/or study-level data too
Explore summary level data of all data. Explore aggregation levels that support the most reasonable data structure and minimize non-independence issues.
#evidence map####
require(maps)
world<-map_data("world")
map<-ggplot() + geom_polygon(data=world, fill="gray50", aes(x=long, y=lat, group=group))
map + geom_point(data=data, aes(x=long, y=lat)) +
labs(x = "longitude", y = "latitude") #render a literal map, i.e. evidence map, of where we study the niche in deserts globally
#add in levels and color code points on map####
map + geom_point(data=data, aes(x=long, y=lat, color = paradigm)) +
scale_color_brewer(palette = "Paired") +
labs(x = "longitude", y = "latitude", color = "")
#aggregation####
se <- function(x){
sd(x)/sqrt(length(x))
}
data.simple <- data %>%
group_by(study.ID, paradigm, technique, measure.success) %>%
summarise(n = n(), mean.lrr = mean(lrr), mean.rii = mean(rii), mean.var = mean(var.es))
main.data <- data %>%
group_by(study.ID, paradigm, intervention, outcome) %>%
summarise(n = n(), mean.lrr = mean(lrr), mean.rii = mean(rii), mean.var = mean(var.es))
#EDA data####
simple.data <- data %>% group_by(study.ID, paradigm, technique, measure.success) %>% summarise(mean.rii = mean(rii), error = se(rii))
simple.data <- na.omit(simple.data)
parad.data <- data %>% group_by(study.ID, paradigm) %>% summarise(mean.rii = mean(rii), error = se(rii))
parad.data <- na.omit(parad.data)
tech.data <- data %>% group_by(study.ID, technique) %>% summarise(mean.rii = mean(rii), error = se(rii))
tech.data <- na.omit(tech.data)
success.data <- data %>% group_by(study.ID, measure.success) %>% summarise(mean.rii = mean(rii), error = se(rii))
success.data <- na.omit(success.data)
#active
active <- data %>%
filter(paradigm == "active")
#viz for aggregation####
ggplot(na.omit(data.simple), aes(technique, n, fill = paradigm)) +
geom_bar(stat = "identity") +
coord_flip() +
scale_fill_brewer(palette = "Paired")
ggplot(na.omit(data.simple), aes(measure.success, n, fill = paradigm)) +
geom_bar(stat = "identity") +
coord_flip() +
scale_fill_brewer(palette = "Paired")
#better
ggplot(main.data, aes(intervention, n, fill = paradigm)) +
geom_bar(stat = "identity") +
coord_flip() +
scale_fill_brewer(palette = "Paired") +
labs(fill = "")
ggplot(main.data, aes(outcome, n, fill = paradigm)) +
geom_bar(stat = "identity") +
coord_flip() +
scale_fill_brewer(palette = "Paired") +
labs(fill = "")
Exploratory data analyses to understand data and QA/QC using Rii.
Meta and conventional statistical models to explore relative efficacy.
Step 5. Synthesis stats
#p-value meta
library(metap)
mdata <- data %>%
filter(paradigm == "active") %>%
filter(!is.na(lrr)) %>%
filter(!is.na(var.es)) %>%
filter(!is.na(n.t)) %>%
filter(!is.na(p)) %>%
filter(!is.na(intervention)) %>%
filter(is.finite(lrr))
schweder(mdata$p)
sumz(p, data = mdata)
## sumz = 52.08743 p = 0
mdata %>%
split(.$intervention) %>%
purrr::map(~sumz(.$p))
## $`carbon amendment`
## sumz = 12.80103 p = 8.089988e-38
##
## $`mycorrhizal inoculation`
## sumz = 10.1732 p = 1.305983e-24
##
## $nutrients
## sumz = 11.64501 p = 1.216527e-31
##
## $planting
## sumz = 35.34839 p = 5.306182e-274
##
## $seeding
## sumz = 29.35404 p = 1.060917e-189
##
## $vegetation
## sumz = 4.627053 p = 1.854529e-06
##
## $`water addition`
## sumz = 13.4316 p = 1.973814e-41
sumlog(mdata$p)
## chisq = 6401.81 with df = 2292 p = 0
#effect-size meta
library(meta)
m <- metagen(lrr, var.es, studlab = intervention, byvar = outcome, data = mdata)
summary(m)
## Number of studies combined: k = 1086
##
## 95%-CI z p-value
## Fixed effect model -0.0063 [-0.0063; -0.0063] -40591712.73 0
## Random effects model 0.0529 [ 0.0094; 0.0964] 2.38 0.0172
##
## Quantifying heterogeneity:
## tau^2 = 0.1848; H = 17034959.73 [17034959.03; 17034960.43]; I^2 = 100.0% [100.0%; 100.0%]
##
## Test of heterogeneity:
## Q d.f. p-value
## 314855990392312640.00 1085 0
##
## Results for subgroups (fixed effect model):
## k 95%-CI
## outcome = native species 244 -0.1445 [-0.1445; -0.1444]
## outcome = soil quality 462 -0.0063 [-0.0063; -0.0063]
## outcome = vegetation and habitat 247 -0.6717 [-0.6717; -0.6717]
## outcome = vertebrate community 7 -0.0208 [-0.0218; -0.0199]
## outcome = invertebrate community 22 0.8279 [ 0.7696; 0.8861]
## outcome = vegetation 104 -1.1042 [-1.1047; -1.1037]
## Q tau^2 I^2
## outcome = native species 1567885581.75 0.2057 100.0%
## outcome = soil quality 314836630773409216.00 0.1848 100.0%
## outcome = vegetation and habitat 13666716955659.36 1.3472 100.0%
## outcome = vertebrate community 14971.11 0.0067 100.0%
## outcome = invertebrate community 58.56 0.0663 64.1%
## outcome = vegetation 1796928.82 0.1222 100.0%
##
## Test for subgroup differences (fixed effect model):
## Q d.f. p-value
## Between groups 5691332250275.88 5 0
## Within groups 314850299060062400.00 1080 0
##
## Results for subgroups (random effects model):
## k 95%-CI
## outcome = native species 244 0.5117 [ 0.3884; 0.6351]
## outcome = soil quality 462 0.2365 [ 0.1810; 0.2919]
## outcome = vegetation and habitat 247 -0.7352 [-1.0015; -0.4689]
## outcome = vertebrate community 7 -0.0080 [-0.1141; 0.0981]
## outcome = invertebrate community 22 0.8366 [ 0.4767; 1.1965]
## outcome = vegetation 104 -0.5343 [-0.6924; -0.3762]
## Q tau^2 I^2
## outcome = native species 1567885581.75 0.2057 100.0%
## outcome = soil quality 314836630773409216.00 0.1848 100.0%
## outcome = vegetation and habitat 13666716955659.36 1.3472 100.0%
## outcome = vertebrate community 14971.11 0.0067 100.0%
## outcome = invertebrate community 58.56 0.0663 64.1%
## outcome = vegetation 1796928.82 0.1222 100.0%
##
## Test for subgroup differences (random effects model):
## Q d.f. p-value
## Between groups 179.21 5 < 0.0001
##
## Details on meta-analytical method:
## - Inverse variance method
## - DerSimonian-Laird estimator for tau^2
funnel(m)
metabias(m)
#forest(m, sortvar = intervention)